Attempt to keep particles on local nodes by comparing the new and old regions and
permuting the solution by picking new regions with the largest number of
cells in the old regions. Gives an unmoved fraction of around 42% for larger
numbers of ranks (compared to 1/nrank for the current scheme).
Doesn't make a huge improvement to runtimes, seems to be around 10% on a
repartition step of EAGLE_25, and we only do that infrequently.